A machine learning model integrating T2* mapping and rest myocardial blood flow sequences for predicting revascularization of individual coronary arteries in stable coronary artery disease

Cardiovascular disease remains the leading cause of death worldwide, with coronary artery disease (CAD) as its most prevalent manifestation. Selection of coronary arteries for revascularization still relies largely on invasive coronary angiography, underscoring the need for effective non-invasive tools. This retrospective single-center study enrolled 843 patients with stable CAD between 2022 and 2025. A development cohort of 468 patients was used for model development, and 375 patients recruited during a later, non-overlapping period formed a same-center temporal validation cohort Four machine learning algorithms were evaluated using six-fold patient-grouped cross-validation, with all coronary-artery observations from the same patient assigned to the same fold. Least absolute shrinkage and selection operator (LASSO) analysis selected four cardiac magnetic resonance (CMR) features and three clinical variables for predicting which coronary arteries underwent in-hospital revascularization. In the development cohort, 723 of 2808 major coronary-artery observations (25.7%) underwent revascularization during hospitalization. The model achieved AUCs of 0.87, 0.94, and 0.94 and precision-recall AUCs of 0.72, 0.90, and 0.76 in the development training cohort, validation set 1, and validation set 2, respectively. The multivariable XGBoost model outperformed individual CMR parameters (AUC, 0.871 vs. 0.815 for T2* mapping and 0.822 for rest myocardial blood flow [rMBF]; all p < 0.05) and single-parameter machine learning models (0.87 vs. 0.84 for T2*-only and 0.86 for rMBF-only models). An XGBoost model integrating clinical variables with conventional CMR parameters, T2* mapping, and rMBF showed discriminative performance for artery-level in-hospital revascularization among patients with stable CAD in the development cohort and same-center temporal validation cohort. Prospective multicenter validation is required before clinical implementation.